New efficient algorithms for multiple change-point detection with kernels - Inria - Institut national de recherche en sciences et technologies du numérique
Pré-Publication, Document De Travail Année : 2016

New efficient algorithms for multiple change-point detection with kernels

Résumé

The present paper focuses on the problem of detecting abrupt changes arising in the full distribution of the observations (not only in the mean or variance). Several statistical approaches based on kernels have been proposed to tackle this problem. Although they enjoy good statistical properties (oracle inequality,.. .), they suffer a high computational cost (in terms of time and memory) which makes them computationally inefficient even with small to medium sample sizes (< 10 4). Our work addresses this computational issue by first describing a new efficient and exact algorithm for kernel multiple change-point detection with an improved worst-case complexity that is quadratic in time and linear in space. It allows to deal with medium size signals (≈ 10 5). Secondly, we also design a faster (but approximate) algorithm based on a low-rank approximation to the Gram matrix, which is linear in time and space. This algorithm can be applied to large-scale signals (n = 10 6). These exact and approximate algorithms have been implemented in R and C for various kernels. The computational and statistical performances of these new algorithms have been assessed through empirical experiments. The runtime of our algorithms is observed to be faster than that of other considered procedures. Finally the simulation results empirically confirm the higher statistical accuracy of kernel-based approaches (and their flexibility) to analyze biological profiles made of DNA copy numbers and allele B frequencies.
Fichier principal
Vignette du fichier
article.pdf (1.51 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-01413230 , version 1 (09-12-2016)
hal-01413230 , version 2 (12-10-2017)

Identifiants

  • HAL Id : hal-01413230 , version 1

Citer

Alain Celisse, Guillemette Marot, Morgane Pierre-Jean, Guillem Rigaill. New efficient algorithms for multiple change-point detection with kernels. 2016. ⟨hal-01413230v1⟩
678 Consultations
1536 Téléchargements

Partager

More